Local structure prediction with local structure-based sequence profiles

MOTIVATION: A large body of experimental and theoretical evidence suggests that local structural determinants are frequently encoded in short segments of protein sequence. Although the local structural information, once recognized, is particularly useful in protein structural and functional analyses, it remains a difficult problem to identify embedded local structural codes based solely on sequence information. RESULTS: In this paper, we describe a local structure prediction method aiming at predicting the backbone structures of nine-residue sequence segments. Two elements are the keys for this local structure prediction procedure. The first key element is the LSBSP1 database, which contains a large number of non-redundant local structure-based sequence profiles for nine-residue structure segments. The second key element is the consensus approach, which identifies a consensus structure from a set of hit structures. The local structure prediction procedure starts by matching a query sequence segment of nine consecutive amino acid residues to all the sequence profiles in the local structure-based sequence profile database (LSBSP1). The consensus structure, which is at the center of the largest structural cluster of the hit structures, is predicted to be the native state structure adopted by the query sequence segment. This local structure prediction method is assessed with a large set of random test protein structures that have not been used in constructing the LSBSP1 database. The benchmark results indicate that the prediction capacities of the novel local structure prediction procedure exceed the prediction capacities of the local backbone structure prediction methods based on the I-sites library by a significant margin. AVAILABILITY: All the computational and assessment procedures have been implemented in the integrated computational system PrISM.1 (Protein Informatics System for Modeling). The system and associated databases for LINUX systems can be downloaded from the website: http://www.columbia.edu/~ay1/.     

Protein backbone angle prediction with machine learning approaches

MOTIVATION: Protein backbone torsion angle prediction provides useful local structural information that goes beyond conventional three-state (alpha, beta and coil) secondary structure predictions. Accurate prediction of protein backbone torsion angles will substantially improve modeling procedures for local structures of protein sequence segments, especially in modeling loop conformations that do not form regular structures as in alpha-helices or beta-strands. RESULTS: We have devised two novel automated methods in protein backbone conformational state prediction: one method is based on support vector machines (SVMs); the other method combines a standard feed-forward back-propagation artificial neural network (NN) with a local structure-based sequence profile database (LSBSP1). Extensive benchmark experiments demonstrate that both methods have improved the prediction accuracy rate over the previously published methods for conformation state prediction when using an alphabet of three or four states. AVAILABILITY: LSBSP1 and the NN algorithm have been implemented in PrISM.1, which is available from www.columbia.edu/~ay1/. SUPPLEMENTARY INFORMATION: Supplementary data for the SVM method can be downloaded from the Website www.cs.columbia.edu/compbio/backbone.     

Prediction of protein local structures and folding fragments based on building-block library

In recent years, protein structure prediction using local structure information has made great progress. In this study, a novel and effective method is developed to predict the local structure and the folding fragments of proteins. First, the proteins with known structures are split into fragments. Second, these fragments, represented by dihedrals, are clustered to produce the building blocks (BBs). Third, an efficient machine learning method is used to predict the local structures of proteins from sequence profiles. Finally, a bi-gram model, trained by an iterated algorithm, is introduced to simulate the interactions of these BBs. For test proteins, the building-block lattice is constructed, which contains all the folding fragments of the proteins. The local structures and the optimal fragments are then obtained by the dynamic programming algorithm. The experiment is performed on a subset of the PDB database with sequence identity less than 25%. The results show that the performance of the method is better than the method that uses only sequence information. When multiple paths are returned, the average classification accuracy of local structures is 72.27% and the average prediction accuracy of local structures is 67.72%, which is a significant improvement in comparison with previous studies. The method can predict not only the local structures but also the folding fragments of proteins. This work is helpful for the ab initio protein structure prediction and especially, the understanding of the folding process of proteins.     

Effects of external interactions on protein sequence-structure relations of beta-trefoil fold

Proteins with symmetric structures are ideal models to investigate the sequence-structure relations. We investigate proteins with beta-trefoil fold and find they have different degrees of sequence symmetries although they show similar symmetric structures. To understand this, we calculate the strength of interactions of the beta-trefoil folds with surrounding environments and find the low degrees of sequence symmetries are often correlated with large external interactions. Our results give an additional confirmation of Anfinsen's thermodynamic hypothesis that protein structures are not only determined by their sequences but also by their surrounding environments. We suggest the external interactions should be considered additionally in protein structure prediction through ab initio folding.     

Assessing a novel approach for predicting local 3D protein structures from sequence

We developed a novel approach for predicting local protein structure from sequence. It relies on the Hybrid Protein Model (HPM), an unsupervised clustering method we previously developed. This model learns three-dimensional protein fragments encoded into a structural alphabet of 16 protein blocks (PBs). Here, we focused on 11-residue fragments encoded as a series of seven PBs and used HPM to cluster them according to their local similarities. We thus built a library of 120 overlapping prototypes (mean fragments from each cluster), with good three-dimensional local approximation, i.e., a mean accuracy of 1.61 A Calpha root-mean-square distance. Our prediction method is intended to optimize the exploitation of the sequence-structure relations deduced from this library of long protein fragments. This was achieved by setting up a system of 120 experts, each defined by logistic regression to optimize the discrimination from sequence of a given prototype relative to the others. For a target sequence window, the experts computed probabilities of sequence-structure compatibility for the prototypes and ranked them, proposing the top scorers as structural candidates. Predictions were defined as successful when a prototype <2.5 A from the true local structure was found among those proposed. Our strategy yielded a prediction rate of 51.2% for an average of 4.2 candidates per sequence window. We also proposed a confidence index to estimate prediction quality. Our approach predicts from sequence alone and will thus provide valuable information for proteins without structural homologs. Candidates will also contribute to global structure prediction by fragment assembly.     

IMPALA: matching a protein sequence against a collection of PSI-BLAST-constructed position-specific score matrices

MOTIVATION: Many studies have shown that database searches using position-specific score matrices (PSSMs) or profiles as queries are more effective at identifying distant protein relationships than are searches that use simple sequences as queries. One popular program for constructing a PSSM and comparing it with a database of sequences is Position-Specific Iterated BLAST (PSI-BLAST). RESULTS: This paper describes a new software package, IMPALA, designed for the complementary procedure of comparing a single query sequence with a database of PSI-BLAST-generated PSSMs. We illustrate the use of IMPALA to search a database of PSSMs for protein folds, and one for protein domains involved in signal transduction. IMPALA's sensitivity to distant biological relationships is very similar to that of PSI-BLAST. However, IMPALA employs a more refined analysis of statistical significance and, unlike PSI-BLAST, guarantees the output of the optimal local alignment by using the rigorous Smith-Waterman algorithm. Also, it is considerably faster when run with a large database of PSSMs than is BLAST or PSI-BLAST when run against the complete non-redundant protein database.     

Analyzing the sequence-structure relationship of a library of local structural prototypes

We present a thorough analysis of the relation between amino acid sequence and local three-dimensional structure in proteins. A library of overlapping local structural prototypes was built using an unsupervised clustering approach called “hybrid protein model” (HPM). The HPM carries out a multiple structural alignment of local folds from a non-redundant protein structure databank encoded into a structural alphabet composed of 16 protein blocks (PBs). Following previous research focusing on the HPM protocol, we have considered gaps in the local structure prototype. This methodology allows to have variable length fragments. Hence, 120 local structure prototypes were obtained. Twenty-five percent of the protein fragments learnt by HPM had gaps.An investigation of tight turns suggested that they are mainly derived from three PB series with precise locations in the HPM. The amino acid information content of the whole conformational classes was tackled by multivariate methods, e.g., canonical correlation analysis. It points out the presence of seven amino acid equivalence classes showing high propensities for preferential local structures. In the same way, definition of “contrast factors” based on sequence-structure properties underline the specificity of certain structural prototypes, e.g., the dependence of Gly or Asn-rich turns to a limited number of PBs, or, the opposition between Pro-rich coils to those enriched in Ser, Thr, Asn and Glu. These results are so useful to analyze the sequence-structure relationships, but could also be used to improve fragment-based method for protein structure prediction from sequence.     

Prediction of protein secondary structure with a reliability score estimated by local sequence clustering

Most algorithms for protein secondary structure prediction are based on machine learning techniques, e.g. neural networks. Good architectures and learning methods have improved the performance continuously. The introduction of profile methods, e.g. PSI-BLAST, has been a major breakthrough in increasing the prediction accuracy to close to 80%. In this paper, a brute-force algorithm is proposed and the reliability of each prediction is estimated by a z-score based on local sequence clustering. This algorithm is intended to perform well for those secondary structures in a protein whose formation is mainly dominated by the neighboring sequences and short-range interactions. A reliability z-score has been defined to estimate the goodness of a putative cluster found for a query sequence in a database. The database for prediction was constructed by experimentally determined, non-redundant protein structures with <25% sequence homology, a list maintained by PDBSELECT. Our test results have shown that this new algorithm, belonging to what is known as nearest neighbor methods, performed very well within the expectation of previous methods and that the reliability z-score as defined was correlated with the reliability of prediction. This led to the possibility of making very accurate predictions for a few selected residues in a protein with an accuracy measure of Q3 > 80%. The further development of this algorithm, and a nucleation mechanism for protein folding are suggested.     

HMMSTR: a hidden Markov model for local sequence-structure correlations in proteins

We describe a hidden Markov model, HMMSTR, for general protein sequence based on the I-sites library of sequence-structure motifs. Unlike the linear hidden Markov models used to model individual protein families, HMMSTR has a highly branched topology and captures recurrent local features of protein sequences and structures that transcend protein family boundaries. The model extends the I-sites library by describing the adjacencies of different sequence-structure motifs as observed in the protein database and, by representing overlapping motifs in a much more compact form, achieves a great reduction in parameters. The HMM attributes a considerably higher probability to coding sequence than does an equivalent dipeptide model, predicts secondary structure with an accuracy of 74.3 %, backbone torsion angles better than any previously reported method and the structural context of beta strands and turns with an accuracy that should be useful for tertiary structure prediction.     

Functional annotation of proteomic sequences based on consensus of sequence and structural analysis

To maximise the assignment of function of the proteins encoded by a genome and to aid the search for novel drug targets, there is an emerging need for sensitive methods of predicting protein function on a genome-wide basis. GeneAtlas is an automated, high-throughput pipeline for the prediction of protein structure and function using sequence similarity detection, homology modelling and fold recognition methods. GeneAtlas is described in detail here. To test GeneAtlas, a 'virtual' genome was used, a subset of PDB structures from the SCOP database, in which the functional relationships are known. GeneAtlas detects additional relationships by building 3D models in comparison with the sequence searching method PSI-BLAST. Functionally related proteins with sequence identity below the twilight zone can be recognised correctly.     

The CATH extended protein-family database: providing structural annotations for genome sequences.

An automatic sequence search and analysis protocol (DomainFinder) based on PSI-BLAST and IMPALA, and using conservative thresholds, has been developed for reliably integrating gene sequences from GenBank into their respective structural families within the CATH domain database (http://www.biochem.ucl.ac.uk/bsm/cath_new). DomainFinder assigns a new gene sequence to a CATH homologous superfamily provided that PSI-BLAST identifies a clear relationship to at least one other Protein Data Bank sequence within that superfamily. This has resulted in an expansion of the CATH protein family database (CATH-PFDB v1.6) from 19,563 domain structures to 176,597 domain sequences. A further 50,000 putative homologous relationships can be identified using less stringent cut-offs and these relationships are maintained within neighbour tables in the CATH Oracle database, pending further evidence of their suggested evolutionary relationship. Analysis of the CATH-PFDB has shown that only 15% of the sequence families are close enough to a known structure for reliable homology modeling. IMPALA/PSI-BLAST profiles have been generated for each of the sequence families in the expanded CATH-PFDB and a web server has been provided so that new sequences may be scanned against the profile library and be assigned to a structure and homologous superfamily.     

Structure-dependent sequence alignment for remotely related proteins

MOTIVATION: The quality of a model structure derived from a comparative modeling procedure is dictated by the accuracy of the predicted sequence-template alignment. As the sequence-template pairs are increasingly remote in sequence relationship, the prediction of the sequence-template alignments becomes increasingly problematic with sequence alignment methods. Structural information of the template, used in connection with the sequence relationship of the sequence-template pair, could significantly improve the accuracy of the sequence-template alignment. In this paper, we describe a sequence-template alignment method that integrates sequence and structural information to enhance the accuracy of sequence-template alignments for distantly related protein pairs. RESULTS: The structure-dependent sequence alignment (SDSA) procedure was optimized for coverage and accuracy on a training set of 412 protein pairs; the structures for each of the training pairs are similar (RMSD< approximately 4A) but the sequence relationship is undetectable (average pair-wise sequence identity = 8%). The optimized SDSA procedure was then applied to extend PSI-BLAST local alignments by calculating the global alignments under the constraint of the residue pairs in the local alignments. This composite alignment procedure was assessed with a testing set of 1421 protein pairs, of which the pair-wise structures are similar (RMSD< approximately 4A) but the sequences are marginally related at best in each pair (average pair-wise sequence identity = 13%). The assessment showed that the composite alignment procedure predicted more aligned residues pairs with an average of 27% increase in correctly aligned residues over the standard PSI-BLAST alignments for the protein pairs in the testing set.     

Increased detection of structural templates using alignments of designed sequences

Protein structure prediction by comparative modeling benefits greatly from the use of multiple sequence alignment information to improve the accuracy of structural template identification and the alignment of target sequences to structural templates. Unfortunately, this benefit is limited to those protein sequences for which at least several natural sequence homologues exist. We show here that the use of large diverse alignments of computationally designed protein sequences confers many of the same benefits as natural sequences in identifying structural templates for comparative modeling targets. A large-scale massively parallelized application of an all-atom protein design algorithm, including a simple model of peptide backbone flexibility, has allowed us to generate 500 diverse, non-native, high-quality sequences for each of 264 protein structures in our test set. PSI-BLAST searches using the sequence profiles generated from the designed sequences ("reverse" BLAST searches) give near-perfect accuracy in identifying true structural homologues of the parent structure, with 54% coverage. In 41 of 49 genomes scanned using reverse BLAST searches, at least one novel structural template (not found by the standard method of PSI-BLAST against PDB) is identified. Further improvements in coverage, through optimizing the scoring function used to design sequences and continued application to new protein structures beyond the test set, will allow this method to mature into a useful strategy for identifying distantly related structural templates.     

Consensus Data Mining (CDM) Protein Secondary Structure Prediction Server: combining GOR V and Fragment Database Mining (FDM)

One of the challenges in protein secondary structure prediction is to overcome the cross-validated 80% prediction accuracy barrier. Here, we propose a novel approach to surpass this barrier. Instead of using a single algorithm that relies on a limited data set for training, we combine two complementary methods having different strengths: Fragment Database Mining (FDM) and GOR V. FDM harnesses the availability of the known protein structures in the Protein Data Bank and provides highly accurate secondary structure predictions when sequentially similar structural fragments are identified. In contrast, the GOR V algorithm is based on information theory, Bayesian statistics, and PSI-BLAST multiple sequence alignments to predict the secondary structure of residues inside a sliding window along a protein chain. A combination of these two different methods benefits from the large number of structures in the PDB and significantly improves the secondary structure prediction accuracy, resulting in Q3 ranging from 67.5 to 93.2%, depending on the availability of highly similar fragments in the Protein Data Bank.     

PISCES: a protein sequence culling server

PISCES is a public server for culling sets of protein sequences from the Protein Data Bank (PDB) by sequence identity and structural quality criteria. PISCES can provide lists culled from the entire PDB or from lists of PDB entries or chains provided by the user. The sequence identities are obtained from PSI-BLAST alignments with position-specific substitution matrices derived from the non-redundant protein sequence database. PISCES therefore provides better lists than servers that use BLAST, which is unable to identify many relationships below 40% sequence identity and often overestimates sequence identity by aligning only well-conserved fragments. PDB sequences are updated weekly. PISCES can also cull non-PDB sequences provided by the user as a list of GenBank identifiers, a FASTA format file, or BLAST/PSI-BLAST output.     

Predicting residue-wise contact orders in proteins by support vector regression

Background  

The residue-wise contact order (RWCO) describes the sequence separations between the residues of interest and its contacting residues in a protein sequence. It is a new kind of one-dimensional protein structure that represents the extent of long-range contacts and is considered as a generalization of contact order. Together with secondary structure, accessible surface area, the B factor, and contact number, RWCO provides comprehensive and indispensable important information to reconstructing the protein three-dimensional structure from a set of one-dimensional structural properties. Accurately predicting RWCO values could have many important applications in protein three-dimensional structure prediction and protein folding rate prediction, and give deep insights into protein sequence-structure relationships.     

A structural alphabet for local protein structures: improved prediction methods

Three-dimensional protein structures can be described with a library of 3D fragments that define a structural alphabet. We have previously proposed such an alphabet, composed of 16 patterns of five consecutive amino acids, called Protein Blocks (PBs). These PBs have been used to describe protein backbones and to predict local structures from protein sequences. The Q16 prediction rate reaches 40.7% with an optimization procedure. This article examines two aspects of PBs. First, we determine the effect of the enlargement of databanks on their definition. The results show that the geometrical features of the different PBs are preserved (local RMSD value equal to 0.41 A on average) and sequence-structure specificities reinforced when databanks are enlarged. Second, we improve the methods for optimizing PB predictions from sequences, revisiting the optimization procedure and exploring different local prediction strategies. Use of a statistical optimization procedure for the sequence-local structure relation improves prediction accuracy by 8% (Q16 = 48.7%). Better recognition of repetitive structures occurs without losing the prediction efficiency of the other local folds. Adding secondary structure prediction improved the accuracy of Q16 by only 1%. An entropy index (Neq), strongly related to the RMSD value of the difference between predicted PBs and true local structures, is proposed to estimate prediction quality. The Neq is linearly correlated with the Q16 prediction rate distributions, computed for a large set of proteins. An "expected" prediction rate QE16 is deduced with a mean error of 5%.     

Homology-based modeling of 3D structures of protein-protein complexes using alignments of modified sequence profiles

Customary practice in predicting 3D structures of protein-protein complexes is employment of various docking methods when the structures of separate monomers are known a priori. The alternative approach, i.e. the template-based prediction with pure sequence information as a starting point, is still considered as being inferior mostly due to presumption that the pool of available structures of protein-protein complexes, which can serve as putative templates, is not sufficiently large. Recently, however, several labs have developed databases containing thousands of 3D structures of protein-protein complexes, which enable statistically reliable testing of homology-based algorithms. In this paper we report the results on homology-based modeling of 3D structures of protein complexes using alignments of modified sequence profiles. The method, called HOMology-BAsed COmplex Prediction (HOMBACOP), has two distinctive features: (I) extra weight on aligning interfacial residues in the dynamical programming algorithm, and (II) increased gap penalties for the interfacial segments. The method was tested against our recently developed ProtCom database and against the Boston University protein-protein BENCHMARK. In both cases, models generated were compared to the models built on basis of customarily protein structure initiative (PSI)-BLAST sequence alignments. It was found that existence of homologous (by the means of PSI-BLAST) templates (44% of cases) enables both methods to produce models of good quality, with the profiles method outperforming the PSI-BLAST models (with respect to the percentage of correctly predicted residues on the complex interface and fraction of native interfacial contacts). The models were evaluated according to the CAPRI assessment criteria and about two thirds of the models were found to fall into acceptable and medium-quality categories. The same comparison of a larger set of 463 protein complexes showed again that profiles generate better models. We further demonstrate, using our ProtCom database, the suitability of the profile alignment algorithm in detecting remote homologues between query and template sequences, where the PSI-BLAST method fails.     

Detecting remote evolutionary relationships among proteins by large-scale semantic embedding

Virtually every molecular biologist has searched a protein or DNA sequence database to find sequences that are evolutionarily related to a given query. Pairwise sequence comparison methods--i.e., measures of similarity between query and target sequences--provide the engine for sequence database search and have been the subject of 30 years of computational research. For the difficult problem of detecting remote evolutionary relationships between protein sequences, the most successful pairwise comparison methods involve building local models (e.g., profile hidden Markov models) of protein sequences. However, recent work in massive data domains like web search and natural language processing demonstrate the advantage of exploiting the global structure of the data space. Motivated by this work, we present a large-scale algorithm called ProtEmbed, which learns an embedding of protein sequences into a low-dimensional "semantic space." Evolutionarily related proteins are embedded in close proximity, and additional pieces of evidence, such as 3D structural similarity or class labels, can be incorporated into the learning process. We find that ProtEmbed achieves superior accuracy to widely used pairwise sequence methods like PSI-BLAST and HHSearch for remote homology detection; it also outperforms our previous RankProp algorithm, which incorporates global structure in the form of a protein similarity network. Finally, the ProtEmbed embedding space can be visualized, both at the global level and local to a given query, yielding intuition about the structure of protein sequence space.     

Use of residue pairs in protein sequence-sequence and sequence-structure alignments

Two new sets of scoring matrices are introduced: H2 for the protein sequence comparison and T2 for the protein sequence-structure correlation. Each element of H2 or T2 measures the frequency with which a pair of amino acid types in one protein, k-residues apart in the sequence, is aligned with another pair of residues, of given amino acid types (for H2) or in given structural states (for T2), in other structurally homologous proteins. There are four types, corresponding to the k-values of 1 to 4, for both H2 and T2. These matrices were set up using a large number of structurally homologous protein pairs, with little sequence homology between the pair, that were recently generated using the structure comparison program SHEBA. The two scoring matrices were incorporated into the main body of the sequence alignment program SSEARCH in the FASTA package and tested in a fold recognition setting in which a set of 107 test sequences were aligned to each of a panel of 3,539 domains that represent all known protein structures. Six procedures were tested; the straight Smith-Waterman (SW) and FASTA procedures, which used the Blosum62 single residue type substitution matrix; BLAST and PSI-BLAST procedures, which also used the Blosum62 matrix; PASH, which used Blosum62 and H2 matrices; and PASSC, which used Blosum62, H2, and T2 matrices. All procedures gave similar results when the probe and target sequences had greater than 30% sequence identity. However, when the sequence identity was below 30%, a similar structure could be found for more sequences using PASSC than using any other procedure. PASH and PSI-BLAST gave the next best results.